OpenAI backs four new California AI-safety bills, urges Congress to act

OpenAI's endorsement of four California bills marks a shift from earlier positions, which the company itself acknowledges by saying it did not endorse some of them in the past and is now supporting them after reconsidering in light of the recent jump in capabilities. That change puts the company alongside a broader state-level effort: it has already supported California's SB 53, New York's RAISE Act, Illinois's SB 315, and independent audits in Massachusetts frontier-safety legislation. OpenAI describes this approach as reverse federalism, where states create a de facto national baseline that Congress can ultimately codify.

The stakes are framed around a closing policy window, a phrase OpenAI borrows from Greg Brockman's defenders-window idea. The company argues that growth in AI capabilities is now outpacing the institutions responsible for governing it, and that the four bills address complementary parts of the safety ecosystem — independent assessment, auditor standards, youth protections, and biological safeguards. OpenAI also points to Astra, its internal research on AI agents completing tasks that would take skilled researchers several days, and an essay by Chief Scientist Jakub Pachocki calling for extreme caution, as evidence that policy needs to move faster.

What the outcome hinges on is whether Congress acts before it adjourns, and whether the states' safeguards converge into a baseline that federal lawmakers can adopt. For frontier labs, the practical question is whether any new federal framework stays targeted at the handful of well-resourced developers rather than becoming, as OpenAI warns, open-weights policy by another name. For smaller developers and researchers, the body says the obligations should be proportionate to capabilities and risks, so the scope of any national framework may be the most consequential detail still to be settled.